Does python-memcached support consistent hashing & binary protocol?

StackOverflow https://stackoverflow.com/questions/2616610

Question

Python-memcached is the official supported memcached driver for Django.

Does it support

  1. Consistent hashing
  2. Binary protocol

If it does, how do I use those features within Django? I couldn't find any documentation.

Was it helpful?

Solution

Looking at the _get_server method on python-memcached v1.45, it seems it doesn't use consistent hashing, but a simple hash % len(buckets).

Same goes for binary protocol, python-memcache uses, as far as I can see in the source, only text commands.

OTHER TIPS

You might be able to use this: http://amix.dk/blog/post/19370

It encapsulates python-memcache's Client class so keys are distributed using consistent hashing.

EDIT- I'm digging in python-memcached 1.4.5 source code, and it looks like it might actually support consistent hashing. Relevant code:

from binascii import crc32   # zlib version is not cross-platform
def cmemcache_hash(key):
    return((((crc32(key) & 0xffffffff) >> 16) & 0x7fff) or 1)
serverHashFunction = cmemcache_hash

-- SNIP --

def _get_server(self, key):
    if isinstance(key, tuple):
        serverhash, key = key
    else:
        serverhash = serverHashFunction(key)

    for i in range(Client._SERVER_RETRIES):
        server = self.buckets[serverhash % len(self.buckets)]
        if server.connect():
            #print "(using server %s)" % server,
            return server, key
        serverhash = serverHashFunction(str(serverhash) + str(i))
    return None, None

Based on this code, it looks like it does implement the algorithm, unless cmemcache_hash is not a meaningful name and that is not the real algorithm. (the now retired cmemcache does consistent hashing)

But I think the OP is referring to more "resilient" consistent hashing, e.g. libketama. I don't think there's a drop in solution out there for that out there, looks like you need to roll up your sleeves compile/install a more advanced memcached lib like pylibmc, and write a custom Django backend that uses that instead of python-memcached.

Anyhow, in either case, some remapping of keys will occur when you add/remove buckets to the pool (even with libketama, just less than with the other algorithms)

Please check this sample python implementation of consistent hashing.

implementation principal : imagine a continnum circle with a number of replicated server points spread across it. When we add a new server, 1/n of the total cache keys will be lost

 '''consistent_hashing.py is a simple demonstration of consistent
hashing.'''

import bisect
import hashlib

class ConsistentHash:
  '''

  To imagine it is like a continnum circle with a number of replicated
  server points spread across it. When we add a new server, 1/n of the total
  cache keys will be lost. 

  consistentHash(n,r) creates a consistent hash object for a 
  cluster of size n, using r replicas. 

  It has three attributes. num_machines and num_replics are
  self-explanatory.  hash_tuples is a list of tuples (j,k,hash), 
  where j ranges over machine numbers (0...n-1), k ranges over 
  replicas (0...r-1), and hash is the corresponding hash value, 
  in the range [0,1).  The tuples are sorted by increasing hash 
  value.

  The class has a single instance method, get_machine(key), which
  returns the number of the machine to which key should be 
  mapped.'''
  def __init__(self,replicas=1):
      self.num_replicas = replicas

  def setup_servers(self,servers=None):
    hash_tuples = [(index,k,my_hash(str(index)+"_"+str(k))) \
               for index,server in enumerate(servers)
               for k in range(int(self.num_replicas) * int(server.weight)) ]
    self.hash_tuples=self.sort(hash_tuples);

  def sort(self,hash_tuples):
    '''Sort the hash tuples based on just the hash values   '''
    hash_tuples.sort(lambda x,y: cmp(x[2],y[2]))
    return hash_tuples

  def add_machine(self,server,siz):
    '''This mathod adds a new machine. Then it updates the server hash
     in the continuum circle '''
    newPoints=[(siz,k,my_hash(str(siz)+"_"+str(k))) \
                   for k in range(self.num_replicas*server.weight)]
    self.hash_tuples.extend(newPoints)
    self.hash_tuples=self.sort(self.hash_tuples);



  def get_machine(self,key):
    '''Returns the number of the machine which key gets sent to.'''
    h = my_hash(key)
    # edge case where we cycle past hash value of 1 and back to 0.
    if h > self.hash_tuples[-1][2]: return self.hash_tuples[0][0]
    hash_values = map(lambda x: x[2],self.hash_tuples)
    index = bisect.bisect_left(hash_values,h)
    return self.hash_tuples[index][0]

def my_hash(key):
  '''my_hash(key) returns a hash in the range [0,1).'''
  return (int(hashlib.md5(key).hexdigest(),16) % 1000000)/1000000.0

Now vbucket is coming for resolving consistent hashing with least impact in cache misses.

If you want a plug-and-play solution for django, use django-memcached-hashring: https://github.com/jezdez/django-memcached-hashring.

It is an adapter around django.core.cache.backends.memcached.MemcachedCache and the hash_ring library.

I have used Consistent hashing algorithm. The lost keys are 1/n of the total number of keys. This means the successful key fetch will be 6/7 *100 around 85%. here

Licensed under: CC-BY-SA with attribution
Not affiliated with StackOverflow
scroll top